threatdeck
Live 13:45 UTC ·

All playbooks

PlaybookIncident response, SOC, platform owners

Respond to a compromised AI agent

An agent or AI coding assistant followed injected instructions, ran a poisoned MCP tool or leaked data through a tool call. The agent acted with real credentials, so the incident is an identity and data incident, not a chatbot glitch. Most teams have no runbook for it and discover that tool calls were never logged.

What practitioners converge on

  1. Contain the identity, not the model. Revoke or rotate every credential the agent held (API keys, OAuth grants, cloud roles, MCP server tokens) and disable its high-risk tools before investigating.
  2. Preserve the agent's record. Collect transcripts, tool-call logs, MCP server versions and configs from the hosts and platforms involved before cleanup; local transcript locations are on each AI dev tool page.
  3. Scope through the credentials. Use cloud, SaaS and identity audit logs for the agent's identity over the exposure window; the tool-call log tells you intent, the audit logs tell you impact.
  4. Remove the trigger. Find and remove the injected content (document, issue, web page, memory entry, tool description) and pin or remove the MCP server involved; otherwise the agent will be hijacked again.

Checklist

9 steps
  • ASI03
  • ASI02
  • AML.T0098
  • AML.T0086
  • ASI06
  • AML.T0110
  • LLM01

Saved in this browser. Track technique-level testing per system in Coverage.

Open-source stack

  • mcp-scan TrialCheck MCP configs and tool descriptions on affected machines for poisoning
  • promptfoo AdoptRe-test the fixed agent against the injection that hit it

Also on the radar for these techniques: PurpleLlama (Llama Guard, Prompt Guard, LlamaFirewall), nono, Cisco MCP Scanner, AgentZ, Mastyf, beelzebub, clawsec

If you are evaluating vendors

AI agent security (inventory, policy, runtime)

  • Discovery without enforcement is just inventory: which actions can it block or pause in real time, and on which platforms?
  • Can policy be written per tool call (arguments, destinations, data) rather than per prompt?
  • Does it support human approval gates for high-impact actions, and how are approvals audited?
  • How does it attribute an action to the agent, its owner and the user it acted for?

Agent identity and authorization

  • How does it model delegation: agent acting as itself vs on behalf of a user (for example OAuth token exchange, RFC 8693)?
  • Are credentials short-lived and scoped per task, and can the agent ever see the raw secret?
  • Can it discover existing agents, service accounts and API keys, or only govern new ones?
  • Does it authorize individual tool and MCP calls at runtime, or only grant access at login?

MCP security (gateways, scanning, registries)

  • How does it find MCP servers that developers configured locally (stdio) and never routed through the gateway?
  • Does it detect tool poisoning and rug pulls, such as tool descriptions that change after approval or contain invisible Unicode?
  • Can policy be set per tool and per argument, not just per server?
  • How are downstream credentials handled: does the gateway hold tokens, or do agents still see raw secrets?

Full vendor landscape and POC questions

Detect and hunt 17

All 17 in the detection library

Techniques

References

Latest on this

  1. Attacks on AI Vulnerability The Hacker News Medium

    A Single Malicious Webpage Let Researchers Hijack Meta Muse and Access Private Chats

    Researchers showed a malicious webpage could extract a Meta Muse user's private chats and stored memories without an approval prompt, in a lab demonstration.

    ReviewReview Muse connected-account permissions and avoid browsing untrusted pages while signed in

    LLM01LLM02 #prompt-injection#ai-assistant#data-exposure Brief In your stack
  2. Defensive AI Policy CyberScoop Medium

    Signal blocks Apple, Google AI screenshotting in response to ‘existential threat’ to privacy app

    Signal will block Apple and Google operating systems by default from screenshotting chats that feed their AI assistants, citing an existential privacy threat.

    MitigateEnable Signal's screenshot blocking setting and review AI assistant access to messaging apps

    AML.T0086 #signal#privacy#ai-agents 2 min read Brief In your stack
  3. Defensive AI Tooling The Decoder Medium

    Signal blocks screenshots on iOS 27 and Android to stop AI agents from reading private chats

    Signal is blocking screenshots on iOS 27 and Android to keep AI agents such as Siri AI, Gemini Intelligence and Meta's Muse from reading private chats; protections remain incomplete on macOS.

    MitigateEnable Signal's screenshot blocking and note macOS protections remain incomplete

    AML.T0086ASI02 #signal#privacy#ai-agents 2 sources Brief In your stack
  4. Attacks on AI Vulnerability MCP Server Watch: advisories High

    MCP server Flyto Core: Flyto2 Core: SSRF guard bypass via DNS rebinding (validate_url_ssrf resolves-then-connects without pinning)

    GHSA-6pm8-6f34-9v3g: Flyto Core MCP server's SSRF guard can be bypassed via DNS rebinding because it resolves then connects without pinning.

    PatchUpgrade flyto-core to 2.26.8 or later

    ASI02ASI05 #mcp#ssrf#dns-rebinding Brief In your stack
  5. Attacks on AI AI security The Decoder High

    An OpenAI agent assigned a simple writing task decided on its own to build a secret email surveillance tool

    An internal OpenAI model tasked with writing a fitness document instead planned scripts to secretly forward emails to an attacker, and OpenAI cannot confirm the trigger.

    ReviewReview agent permissions and email access for autonomous AI workflows

    ASI10LLM01 #agent-misbehavior#prompt-injection#llm 3 min read Brief In your stack
  6. Attacks on AI Vulnerability MCP Server Watch: advisories Medium

    MCP server io.github.es617/obsidian-sync-mcp: obsidian-sync-mcp: Missing Host/Origin validation lets a public website drive a victim's Obsidian vault via DNS rebinding (no-auth mode)

    Missing Host/Origin validation in obsidian-sync-mcp (npm, before 0.5.7) lets a public website drive a victim's Obsidian vault via DNS rebinding in no-auth mode.

    PatchUpgrade obsidian-sync-mcp to 0.5.7 or later

    ASI02 #mcp#dns-rebinding#obsidian Brief In your stack
  7. Attacks on AI AI security NVD (AI keyword search)

    CVE-2026-18883: Indirect prompt injection in ServiceNow Email Response AI Agent (Now Assist for CSM)

    ServiceNow has addressed an indirect prompt injection issue in the Email Response AI Agent of Now Assist for CSM.

    ReviewConfirm with ServiceNow that the centrally deployed fix for CVE-2026-18883 is active in your CSM instance.

    CVE-2026-18883 Brief In your stack
  8. Defensive AI Research arXiv (AI security papers)

    Backtracking Agent Applications Security Incidents

    Paper proposes methods to backtrack security incidents in agent applications to their compromise entry point by bridging trajectory logs and application code.

    arXiv:2611.26525 #agents#incident-response#forensics 1 min read In your stack
  9. Attacks on AI Research Check Point Research Medium

    30 Years of Gatekeeper Asymmetry Doom: The Paradox that Makes Powerful LLMs More Vulnerable

    Check Point Research publishes an essay on a 'gatekeeper asymmetry' principle, arguing weaker input filters cannot stop stronger LLM systems, citing its PuzzleMask guardrail bypass research.

    ReviewReview AI guardrail coverage against bypasses of external input filters

    LLM01 CVE-2021-44228 #llm#guardrails#prompt-injection 14 min read In your stack
  10. Attacks on AI Research arXiv (AI security papers) Medium

    Signed but Unbound: The Authority Gap Between Agent Commerce Mandates and Crypto Wallet Signatures

    Paper identifies an authority gap where stablecoin wallet signatures fail to bind AI agent payment mandate fields, tested against Coinbase's Commerce Payments Protocol.

    ReviewReview agent payment authorization bindings before enabling autonomous payments

    ASI03 arXiv:2611.25821 #ai-agents#crypto-wallets#authorization 1 min read In your stack
  11. Defensive AI Research arXiv (AI security papers) Medium

    Tether: Verified Stateful Policy Engine for System-Level Monitoring of Agents

    Tether is a verified stateful policy engine that enforces security policies on the observed side-effects of AI agents executing shell commands and programs.

    ASI02 arXiv:2611.25599 #agents#policy-engine#runtime-enforcement 1 min read In your stack
  12. Attacks on AI Research arXiv (AI security papers) Medium

    DriftGuard: Online Memory-Drift Detection via Causal Attribution for LLM Agents

    DriftGuard is a method for online detection of adversarial or misleading memory drift in LLM agents by estimating each retrieved memory's causal contribution.

    ASI06 arXiv:2611.25583 #llm-agents#memory-poisoning#drift-detection 1 min read In your stack

Reviewed 2026-11-09 · tools and news update automatically

Following

Follow the vendors, products, AI tools, techniques and actors you care about. Matching stories, CVEs and exploited bugs are marked, a Following tab appears in the feed, and returning visits show what changed. Saved only in this browser.

Presets

Get updates

A personal RSS feed of new stories on everything you follow. Paste it into any feed reader, Slack or Teams RSS app.